Search Images Maps Play YouTube News Gmail Drive More »
Sign in
Screen reader users: click this link for accessible mode. Accessible mode has the same essential features but works better with your reader.

Patents

  1. Advanced Patent Search
Publication numberUS4791676 A
Publication typeGrant
Application numberUS 07/148,169
Publication dateDec 13, 1988
Filing dateJan 22, 1988
Priority dateSep 12, 1985
Fee statusLapsed
Publication number07148169, 148169, US 4791676 A, US 4791676A, US-A-4791676, US4791676 A, US4791676A
InventorsMyron D. Flickner, Kottappuram M. Ali Mohiuddin
Original AssigneeInternational Business Machines Corporation
Export CitationBiBTeX, EndNote, RefMan
External Links: USPTO, USPTO Assignment, Espacenet
Method and means for efficiently handling boundary conditions in connected component labeling
US 4791676 A
Abstract
A method and means for analyzing and labeling picture elements ( pixels) in a raster-scanned image includes utilizing a fixed pixel labeling priority rule that takes into account near-neighbor frame border pixels that are not included in the currently-viewed portion of the image by substituting the inverse of the value of the current image pixel being labeled in place of undefined border pixels outside the viewed image portion.
Images(4)
Previous page
Next page
Claims(13)
We claim:
1. In a method of labeling the components of a black/white raster-scanned image frame in which the color of a component to be labeled is analyzed with respect to the colors of its causal neighbors contained in a fixed-sized window and components are labeled sequentially in response to a color valve match analysis of a sequence of windows, the improvement comprising:
identifying a frame boundary in a fixed-sized window, said frame boundary including undefined neighbor components of the component being analyzed for labeling; and
in response to said identification and to the color value of the component being analyzed, substituting the inverse of the color value of the component being analyzed in the places of said undefined neighbor components.
2. The method of claim 1 wherein said step of identifying includes identifying a first boundary, said first boundary including a sequence of undefined components outside of said image frame and adjacent an edge of said image frame.
3. The method of claim 2 wherein said step of identifying further includes identifying an extreme boundary, said extreme boundary including a second sequence of undefined components outside of said image from and separated by said first boundary from said edge.
4. The method of claim 1 wherein said step of identifying includes identifying respective intersecting boundaries in said window.
5. In a system for sequentially labeling the pixels of a black/white raster-scanned frame image in response to the color values of pixels contained in a sequence of fixed-sized windows defined in said image, the improvement comprising:
a boundary detection circuit which responds to said raster-scanned image frame by providing a signal identifying a frame boundary contained in one of said sequence of fixed-sized windows, said frame boundary including undefined boundary pixels contained in said window; and
a pixel value selection circuit which responds to said signal and to said color values of said pixels by assigning to said undefined pixels the inverse color value of a pixel contained in said fixed-sized window and which is next to be labeled.
6. The improvement of claim 5 wherein said frame boundary includes a sequence of undefined pixels outside of and adjacent to an edge of said raster-scanned image.
7. The improvement of claim 5 wherein said frame boundary is an extreme frame boundary, said extreme frame boundary including a first column of undefined pixels outside of said raster-scanned image and separated from an edge of said raster-scanned image by a near frame boundary, said near frame boundary including a second column of undefined pixels outside of and abutting an edge of said raster-scanned image.
8. In the analysis of a fixed-sized frame of black/white raster-scanned images, a method for connected component labeling of said frame in a sequential manner utilizing a predetermined fixed priority rule, comprising the steps:
analyzing the connectivity of each component in each of the black/white images by assessing a color value match of each component with respect to its causal neighbors within a fixed-sized window;
labeling each analyzed component based upon its connectivity with a connected group of components within the frame, any given component being labeled according to the fixed priority rule in the event that at least two of its causal neighbors have the same color value.
identifying a frame boundary in a fixed-size frame, said frame boundary including undefined neighbor components of a component being labeled; and
in response to said identification and to the color value of the component being labeled, substituting the inverse of the color value of said component being labeled in the places of said undefined neighbor components.
9. The method of claim 8 wherein said components are pixels occurring in a raster-scanned sequence synchronized by a pixel clock signal having a predetermined pixel frequency and said steps are performed for each of said pixels within the cycle of said pixel clock signal during which the pixel occurs.
10. An apparatus for sequentially analyzing the connectivity of pixels in a raster-scanned image frame according to a predetermined labeling priority rule, comprising:
window register means responsive to a sequence of color-valued pixels that form a raster-scanned image frame for associating each pixel in said sequence to causal pixel neighbors by a fixed-sized window containing the pixel and its causal neighbors;
labeling means responsive to each fixed-sized window for analyzing the connectivity of each pixel to other pixels in the sequence based upon a color value match of the pixel with respect to its causal neighbors and for labeling each pixel based upon its connectivity with a connected group of components within the frame;
a frame boundary means responsive to said image frame for recognizing a frame boundary contained in a fixed-sized window, said frame boundary including undefined neighbor pixels of a pixel being labeled; and
value substitution means connected to said window means, said frame boundary means, and said labeling means and responsive to said frame boundary recognition and to the color value of the pixel labeled for substituting the inverse of the color value of said pixel being labeled in the places of its undefined pixel neighbors.
11. The apparatus of claim 10 wherein said labeling means includes an addressable storage device having pixel label selection signals stored at locations addressed by color values of pixels in said fixed-size windows.
12. The apparatus of claim 11 wherein said fixed-size window pixel color values include said inverted values substituted in the places of undefined pixels contained in said fixed-size windows adjacent the boundaries of said image frame.
13. The apparatus of claim 10 wherein said value substitution means includes programmable logic circuitry.
Description

This is a continuation of application Ser. No. 775,460, filed Sept. 12, 1985, now abandoned.

BACKGROUND OF THE INVENTION

The invention relates to the analysis and labeling of individual picture elements (pixels) of a raster-scanned image by utilizing a fixed labeling priority rule, and more particularly to the labeling of pixels positioned adjacent the borders of a raster-scanned image frame.

Automated recognition of objects in raster-scanned images is aided by the technique of connectivity analysis in which the pixels forming a particular object in the scanned image are assigned the same label. An operation for labeling image object pixels was originally set forth by Rosenfeld and Pfaltz in "Sequential Operations and Digital Picture Processing," Journal Of ACM, vol. 14, no. 4, October 1966, pp. 471-494. U.S. Pat. No. 4,183,013 to Agrawala et al. exemplifies a pixel labeling scheme.

With respect to pixel labeling, the pixels forming a given object are said to be connected, and the Rosenfeld reference teaches that the labeling of image pixels involves the two steps of analyzing the connectivity of pixels in the image and labeling of pixels in analyzed image regions.

Pixel connectivity analysis is typically handled by the examination of a neighborhood of pixels surrounding a pixel to be labeled, with the currently-analyzed pixel assigned the value of one of the neighborhood pixels according to a predetermined priority scheme. Using the prioritized neighborhood, the pixels of a raster-scanned image can be analyzed and labeled in the sequence in which they are generated.

Typically, the viewed portion of a raster-scanned image consists of a window that is framed over a larger pixel matrix, so that the viewed portion is surrounded by "undefined pixels" not actually belonging to the viewed image. These undefined pixels are not labeled. However, when the currently-analyzed pixel is adjacent the edge of the viewed portion, the neighborhood employed for its analysis and labeling includes undefined pixels lying outside the framed image window.

When analyzing and labeling pixels on the edge of the viewed portion of a scanned image, previous label selection priority schemes are forced to maintain a set of specialized priority rules, each employed at a respective one of the boundaries of the viewed image. This makes the labeling process difficult, with the challenge being to find an efficient, simplified way of accommodating image boundary conditions so that the undefined pixels do not lead to ambiguous results in component labeling or to complex labeling algorithms.

The ambiguity arises in connectivity labeling schemes from the fact that undefined pixels contain no image information that validly relates them to the pixels forming the currently-viewed portion of an image. Assignment of an arbitrary value to undefined pixels can result in the assignment of an incorrect label to an image boundary pixel if the image boundary pixel's label is selected on the basis of an undefined pixel to which the connectivity scheme determines the image pixel is connected.

Many labeling schemes (such as the Agrawala one referenced previously) employ complex software constructs for pixel analysis and labeling; these are relatively slow-acting and inappropriate for use in high-speed sequential processing such as is implemented in pipelined systems.

Accordingly, the principal object of the present invention is to provide for the unambiguous labeling of image boundary pixels during connected pixel labeling of the image in a sequential manner utilizing a fixed priority rule.

An advantage conferred by achieving this objective is that component labeling according to the present invention can be speedily and efficiently done on a sequential basis, making it particularly suited for high speed pipelined hardware implementation.

SUMMARY OF THE INVENTION

The present invention provides, in the analysis of frames of raster-scanned images, a method and means for connected pixel labeling of an image frame in a sequential manner utilizing a fixed labeling priority rule.

The method includes analysis of the connectivity of each pixel in a black/white image frame by sequential examination of the color match of each pixel with respect to its adjacent neighbors within a predetermined neighborhood. Each analyzed pixel is labeled such that the same label is assigned to each pixel within a maximally connected group of pixels within the image frame, with any given pixel being labeled according to a fixed priority rule in the event that at least two causal neighbors within the neighborhood have the same color match. At the image frame boundary neighborhoods including undefined neighbor pixels, the inverse of the value assigned the pixel being labeled is substituted in the places of the undefined neighbor pixels.

The invention is also expressed in the form of an apparatus for analyzing and labeling fixed frames of a raster-scanned image and labeling connected pixels of an image frame in a sequential manner utilizing the same fixed priority rule.

The apparatus includes a pixel labeling window register for defining a fixed size pixel window for each pixel in a sequence of pixels forming a raster-scanned black/white image. A pixel labeling circuit responds to the values of pixels in each window by labeling each pixel such that the same label is assigned to each pixel within a maximally connected group of pixels within an image frame, any given pixel being labeled by the pixel labeling circuit according to a fixed priority rule in the event that at least two causal neighbors in the given pixel's window have the same color match. The apparatus further includes pixel value substitution logic for, when a window defined by the pixel labeling circuit contains undefined pixels, substituting, in the places of the undefined pixels, the inverse of the value assigned the pixel being labeled.

Therefore, the summarized method and apparatus meet the stated objective by removing undefined pixels from the analysis and labeling of a pixel by insuring that they are assigned color values that do not match the value of the pixel being labeled.

Other objectives and advantages of the present invention will become apparent when the following detailed description is read in conjunction with the below-described drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 represents a raster-scanned image containing three image objects and four frames.

FIG. 2 is a magnified view of a portion of a raster-scanned image frame illustrating the matrix arrangement of pixels in the image.

FIG. 3 illustrates a preferred component labeling window used in analyzing and labeling the sequence of pixels forming a raster-scanned image.

FIGS. 4A and 4B illustrate three separate boundary conditions accounted for by the invention in the analysis and labeling of pixels.

FIG. 5 is a block diagram illustrating the use of the invention in a pixel labeling system.

FIG. 6 is a partial block diagram illustrating in greater detail the interconnection of pixel value substitution logic between a set of image window registers and a pixel labeling circuit.

FIGS. 7A and 7B illustrate correct and incorrect labeling of image object pixels.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Refer to FIG. 1 for an understanding of object recognition in a black/white raster-scanned image. In the raster-scanned image of FIG. 1, identification of objects 10, 12, and 14 can be accomplished at a low level of image abstraction by providing the pixels connected to form a respective object with identical labels to differentiate them from pixels forming another, different object.

In this regard, pixels (picture elements) are the elemental image regions that are presented for viewing in an ordered NM matrix format which is represented by the matrix fragment of FIG. 2. The matrix of pixels is formed by scanning a scene in a vertically ordered sequence of horizontal sweeps forming a raster, and quantizing the scene information at regular intervals in each sweep to provide a succession of pixels. A sweep is conventionally referred to as a scanline and an image is reconstructed from a swept scene by presenting the scanlines in the sequence with which they were generated. The separation of each scanline into a sequence of pixels and the presentation of each scanline adjacent the scanline following it naturally imposes the matrix format of FIG. 2. The matrix structure (conventionally scalled a raster-scanned format) has N rows (each corresponding to a respective scanline), each made up of a sequence of M pixels. The matrix thus includes M columns of vertically adjacent pixels and is referred to as an "NM matrix." In the discussion following, whenever a pixel or image matrix is referred to, it is to be considered an NM, raster-scanned matrix of pixels.

Referring to FIG. 1, the objective of an image analysis application is to analyze and identify the pixels of an image according to the respective image objects that they form. To facilitate the identification, each pixel is assigned one of a pair of binary values by comparing it against a preset level. If a pixel meets or exceeds the level it is assigned a color value of 1, indicating that it is connected to other, as yet unidentified pixels, to form an image object. Pixels that fall below the preset level are assigned to make up the image background or holes in objects. Since these pixels do not form part of an image object they are assigned the second binary value 0.

In FIG. 1 all of the pixels forming the background upon which the image objects 10, 12, and 14 are presented are assigned the label H0 by the labeling application process, with the H indicating their pixel value (0) and the subscript indicating their inclusion in the background. The pixels forming the image object 10 are assigned the label I1, while the image pixels of the object 12 are assigned the label I2 to distinguish them from the image pixels of the object 10 that have the same binary value. In these cases, the I portion of the label indicates the original assignment of a 1 to such pixels.

The labeling of pixels associated with the image object 12 (a circular ring) must also include labeling the pixels in the hole of the middle of the ring to distinguish them from the background pixels H0. The hole pixels in the image object 14 are therefore assigned the label H2.

The resolution requirements of many image analysis applications are such that the entire image cannot be encompassed in one field of view. What is often done in such cases is to scan the image in frames, analyze the frames separately, and then combine the results from the different frames to form a composite description of the image. For example, in FIG. 1 the oval-shaped ring (image object 14) has portions in each of four abutting image frames 16, 18, 20, and 22. Although the pixels for the four portions of the object 14 will be initially labeled separately, it is to be understood that the application process analyzing the image of FIG. 1 will have means to piece together the differently-labeled object portions into an integrated image object. Proper merging of the portions of the object 14 requires unambiguous labeling of the pixels in each of its sections. With further particularity, the labeling scheme must insure that the hole pixels of the object 14 are assigned H labels distinguishable from the H label assigned backgroundd pixels in the image of FIG. 1. This characterizes the problem solved by the method and apparatus of the invention.

Refer to FIG. 3 now for an understanding of how connected pixels in an image are labeled by neighborhood analysis. In the invention, the current pixel being analyzed and labeled is denoted by P. A 35 window of pixels surrounding the current pixel P being examined and labeled is used to analyze its connectivity to adjacent pixels.

The 35 pixel labeling window used in pixel labeling has a first component of 8 pixels A-H that immediately surround the current pixel P in a basic 33 pixel square that is framed on each end by a 31 pixel stack. The 35 window is used to establish the connectivity relation between the current pixel P and the image pixels that have already been labeled. In the following portion of the discussion, the pixels in the pixel labeling window are referred to as the "neighbors" of P.

In the labeling scheme, if P has the same value as one of its connected neighbors in the pixel window, then P will be assigned the label of this neighbor. Since there may exist more than one connected neighbor, a priority sequence is followed for selecting the label to be assigned P. The preferred priority sequence for purposes of understanding the invention is: label (A)>label (B)>label (C)>label (D)>label (V)>label (X).

In connectivity analysis, A, B, C, D, V, and W are referred to as P's causal neighbors. Given the general left-to-right and top-to-bottom path of raster image scanning, a relational history of these causal neighbors can be accumulated by the labeling application to provide a reliable basis for making a decision concerning their connectivity to P.

Refer now to FIG. 4A for consideration of a pixel labeling window (PLW) adjacent the left edge of an image frame. Initially, it is to be understood that the frame defining an image to be viewed does not necessarily include all of the pixels in the image, and that undefined pixels lie adjacent the perimeter of the image frame, yet do not belong to the image. Thus, when P is adjacent the edge of the frame, the priority selection scheme will not work accurately. For example, when, as in FIG. 4A, P (defined by the intersection of columnl and rowj) is at the left perimeter of an image frame, the first two columns of the PLW will contain "undefined pixels" (indicated by "*"), while the remaining three columns will all have pixels defined with respect to the image frame being analyzed (indicated by "."). Therefore, the PLW in FIG. 4A falls on the left extreme boundary (LEB). If the PLW is relocated by one pixel to the right in FIG. 4A, then the neighbors U, V, and W will still be undefined; this location is referred to as the left near boundary (LNB). It should be noted that corresponding boundary conditions exist on the right-hand edge of the raster-scanned image; these are referred to as right extreme boundary (REB) and right near boundary (RNB).

In FIG. 4B, P lies at the intersection of the first row of the raster matrix, rowl, and columnh. Therefore, the top row of the PLW contains undefined pixels and is said to lie in the top boundary (TB). A corresponding bottom boundary (BB) of undefined pixels borders the bottom edge of the raster matrix.

In view of the described boundaries, and, assuming that the image frame matrix is larger than or equal to the component labeling window, 15 distinct boundary conditions must be accounted for when analyzing and labeling P. These conditions are given in Table I.

              TABLE I______________________________________BOUNDARY     BOUNDARY CONDITIONCONDITION    DEFINING SIGNALS______________________________________ 1         ##STR1## 2         ##STR2## 3         ##STR3## 4         ##STR4## 5         ##STR5## 6         ##STR6## 7         ##STR7## 8         ##STR8## 9         ##STR9## 10         ##STR10## 11         ##STR11## 12         ##STR12## 13         ##STR13## 14         ##STR14## 15         ##STR15##______________________________________

For example, condition 1 defines the condition where P is adjacent the left extreme boundary and in neither the top left-hand nor the top right-hand corner of the raster matrix. Condition 5 corresponds to P being located in the bottom row (rown) and second column (col2) of the matrix. Condition 13 obtains whenever the PLW is located such that P is in rowl of the raster matrix and the PLW is not adjacent either the left or right extreme or near boundaries. Under condition 15, no pixels in the PLW are undefined.

Returning again to the example of FIG. 4A, the practice of the invention requires that, for each of the boundary conditions 1-14, the inverse of the value of P be assigned arbitrarily to the undefined pixels in the PLW. This will remove the undefined pixels from inclusion in the above-described pixel labeling scheme, and remove any possibility of ambiguity in the labeling of pixels positioned adjacent the periphery of a raster-scanned image frame.

An apparatus for performing the above-described method can be understood with reference to FIG. 5. In FIG. 5 a conventional video control and timing circuit 28 produces a SYNCH signal to synchronize raster scanning and a pixel clock (PCLK) having the frequency of pixel occurrence in a row of a raster-scanned image. A raster-scanned image (RSI) generator 30 conventionally produces an image signal consisting of a stream of quantized, gray-scale image pixels in a standard raster-scanned format, with a new image pixel produced with each cycle of PCLK. Conventional pixel processing and threshold logic 32 compares each pixel in the input pixel stream against a predetermined level to derive the binary characteristic for the pixel. The output of the pixel measurement logic 32 is provided as a stream of binary pixels, each having only an image (1) or hole (0) value.

The pixel stream produced by the pixel measurement logic 32 is fed to a set of pixel labeling window (PLW) registers 40 which frame the PLW for each pixel of the raster-scanned image that is to be labeled. The registers are conventional, serially-connected shift registers which shift the pixels of an NM image frame pixel matrix serially at the frequency of PCLK. The image window registers 40 include a first register 40a having a 15 pixel capacity. The 5-pixel register 40a is connected to a shift register 40b having the capacity to store and serially shift M-5 pixels. These registers 40a and 40b together have the capacity for storing 1 row of M pixels. The register 40b shifts pixels into another 15 pixel register 40c. The register 40c is connected to shift pixels into a register 40d, equivalent in all respects to the register 40b. Finally, the register 40d shifts pixels to a third 15 pixel shift register 40e.

As pixels are shifted serially through the PLW registers 40, the pixels held in locations F, G, and H of register 40a, the pixels in positions V, A, P, and E of register 40c, and the pixels in positions B, C, D, and X of register 40e are provided, once each cycle of PCLK, on the twelve output lines labeled IWP's (corresponding to image window pixels).

The IWP's are fed to a pixel value substitution logic (PVS logic) circuit 43 which detects the current boundary condition and which, if necessary, also substitutes the inverted value of P for the pixels that are undefined under the current boundary condition.

In response to boundary condition 15 of Table I, the PVS logic circuit 43 forwards all of the IWP's obtained from the PLW registers 40 to a pixel labeling (PL) PROM 46. Under each of conditions 1-14, the PVS logic circuit 43 forwards the values of defined pixels together with the inverse value of P in place of the undefined pixels.

The PL PROM 44 is programmed to implement the labeling scheme described above, using the 12 forwarded pixels to address a storage location indicating which of the labels is to be assigned to P.

The operations of the PVS logic circuit 43 and the PL PROM 44 are synchronized through provision of PCLK.

The current boundary condition obtaining for a raster-scanned image is provided by a top/bottom boundary (TBB) detection circuit embracing components 45a, 45b, and 45c, and a right/left boundary detection circuit including components 46a, 46b, and 46c. The top/bottom boundary detection circuit includes a row counter 45a that identifies the number of the current row being scanned by counting up in response to the column counter described below. The current row count from the row counter 45a is compared against a value contained in a presettable image height register (IHR) 45b, the comparison being done by a row count comparator (RCC) 45c. The RCC operates as follows: when the row counter 45a begins counting from 0, the comparator 45c provides the top boundary (TB) signal. The row count is incremented each time the column counter reaches a count of M. Then, when the row counter and IHR values are equal, the comparator 45c provides the bottom boundary (BB) signal. In the preferred embodiment, the IHR 45b is set to N, so that BB is activated when the row count reaches N. The row counter 45a is reset upon reaching N and begins counting again from 0. In this regard, a signal to reset the row counter can be provided from the RCC 45c.

Raster matrix columns are counted by a column counter 45a that is reset at the beginning of each scanline and incremented thereafter by PCLK. The current count of the counter 46a is 5 compared against the width in pixels of the raster-scanned image as represented by a value in an image width register (IWR) 46b. A column count comparator (CCC) 46c compares the output of the counter 46a against the image width available from the register 46b. In keeping with the defined matrix structure, the IHR 46b is set to M and the column counter 46b counts PCLK cycles from 0 to M. The column counter 46b is reset when it reaches M; each time the column counter is reset, the row counter is incremented.

The comparator 46c compares the preset count in the register 46b with the current column count in the counter 46a and at the appropriate time provides the signals LEB, LNB, RNB, or REB.

It should be evident that the boundary signals TB, BB, LEB, LNB, RNB, and REB are all synchronized by PCLK to the shifting of pixels through the PLW registers 40. The boundary signals indicate the presence in the PLW registers of undefined pixels from previous pixel rows in the same frame or the immediately preceding or succeeding frame. Thus, when TB and LEB are activated, indicating boundary condition 1, undefined pixels, corresponding to the pixels forming the bottom row of the preceding interation of the current frame, will occupy the PLW regiters 40 from position A in register 40c through position W in register 40e. The values for M defined pixels (corresponding to the first row of the present scanned iteration of the current frame) will be contained in the PLW registers from storage position P backward through position H in register 40a. As another example, RNB will indicate when the first two pixels of rowi+2 have been shifted into locations F and Z in register 40a. Rowi+1 pixels will be held from location G in register 40a through E in register 40c; rowi pixels from P in register 40c through D in register 40e; and rowi-1 pixels from C through W in register 40e. In this example, the RNB boundary signal will cause the PVS logic circuit 43 to substitute the inverted value of P for the pixel values stored at positions Z, F, Y, E, D, and X of the PLW registers 40. The valid and substituted pixel values are provided, in the proper order, by the PVS logic circuit 43 to the address (A) port of the PL PROM 44.

A pixel label selection signal corresponding to the label selected according to the above-described algorithm is stored at the location addressed by the original and substituted values forwarded by the PVS logic circuit 43 to the PL PROM 44. It should be evident that, with the inverted value of P substituted for the invalid pixels, the invalid pixels will not affect the selection of a label for P.

The interconnection of the PVS logic circuit 43 between the PLW registers 40 and the PL PROM 44 is illustrated in greater detail in FIG. 6. In the preferred embodiment, the inventors have embodied the functions of the PVS logic circuit 43 in a programmable logic array (PLA) that receives as inputs the six boundary signals and the twelve IWP values from the PLW registers 40a, 40c, and 40e. The boundary signals are provided through a select (S) port of the logic circuit 43; the IWP values from the PLW registers are provided through an input (I) port; and the PCLK signal is fed through a clock port.

Programmable array logic is well understood in the art. A programmable logic device such as the PVS logic circuit 43 is capable of being programmed to perform functions in response to a varying set of input conditions. In the invention, the PVS logic circuit 43 is programmed to execute in response to the boundary signals by operating on the IWP's in a manner corresponding to the pseudo-Pascal code presented in Table II.

              TABLE II______________________________________ PIXEL VALUE SUBSTITUTION PROCEDURE______________________________________/* Initially assume condition 15 exists (no boundarysignals active) and make the following directassignments */       Aout : = Ain ;       Bout : = Bin ;       Cout : = Cin ;       Eout : = Din ;       Fout : = Ein ;       Gout : = Fin ;       Hout : = Hin ;       Vout : = Vin ;       Xout : = Xin ;       Yout : = Yin ;       Pout : = Pin ;/* Now, as boundary signals are activated, substituteappropriate pixels in the following way */If LEB thenbegin       Vout : = -P;       Bout : = -P;       Aout : = -P;       Hout : = -Pendelse if LNB then       Vout : = -P;if RNB thenbegin       X.sub. out : = -P;       Yout : = -Pendelse if REB thenbegin       Xout : = -P;       Yout : = -P;       Dout : = -P;       Eout : = -P;       Fout : = -Pend;if TB thenbegin       Bout : = -P;       Cout : = -P;       Dout : = -P;       Xout : = -Pend;if BB thenbegin       Hout : = -P;       Gout : = -P;       Fout : = -Pend.______________________________________

In Table II, the PVS logic 43 initially assumes that condition 15 exists, that is that P is not adjacent any of the image boundaries; in this case, the PVS logic circuit 43 will simply assign to the output pixels (Aout -Pout) the pixel values for the corresponding IWP's provided by the PLW registers 40 (Ain -Pin).

Next, as the boundary signals provided by the comparators 45c and 46c are activated, that is as pixels adjacent the above-defined boundaries of the raster-scanned image are labeled, the undefined pixels in the PLW are neutralized by substitution of the inverted P value for the values of corresponding invalid pixels. In this regard, when P is on the left extreme boundary, LEB is activated and the portion of the pseudo-code listing headed "IF LEB then" causes -P, the inverted value of P, to be substituted into the window neighbors overlaying undefined pixels. In the left extreme boundary case, these neighbors are A, B, H, and V.

The procedure of Table II also provides for label selection of corner pixels by permitting assignment of the inverted P value in the extent that a vertical and a horizontal boundary signal are concurrently activated.

Alternative approaches to altering the labeling algorithm embedded in the PL PROM 44 are possible in light of the above teachings. For example, it would be possible to encode the six BOUNDARY SIGNALS into a 4-bit boundary condition signal and feed the boundary condition signal directly to the PL PROM 44. This, however, would require increasing the size of the PROM by a factor of 16.

Another, erroneous approach, deceptively attractive in its simplicity, would be to pad the image frame with undefined pixels having a predetermined value, for example 0 or 1. Such padding would be made possible by altering the Table II procedure to assign the predetermined value to the undefined pixels indicated by a respective BOUNDARY SIGNAL. Such padding would effectively surround the border of the raster-scanned image with an enclosure of identically-valued undefined pixels. However, as illustrated in FIG. 7A and 7B this could lead to incorrect labeling of valid image pixels. In FIG. 7A a labeled image consisting of an object and four holes is illustrated. Proper labeling will assign the connected pixels of the object the pixel label I1, while the pixels of each of the holes would bear labels significating that the holes are different. If the undefined pixels surrounding the FIG. 7A image were arbitrarily assigned the value 0, and the component labeling algorithm described above were followed, the result would be to label all the holes identically as illustrated in FIG. 7B.

It is therefore manifest that the disclosed invention enhances image inspection applications employing connectivity analysis by component (pixel) labeling. The enhancement results from responding dynamically to the changing border of a raster-scanned image and acting to suppress the potentially ambiguous effect of undefined boundary pixels in the labeling of image pixels adjacent the border. It should be evident that the proper labeling of boundary pixels according to the invention enables the analysis of images that are larger than an available viewing frame. The solution to the boundary value problem provided by the invention limits the size of memory required to support a component labeling system by minimizing the size of the pixel labeling PROM 44.

Finally, the apparatus of the invention, with proper selection of circuitry, can perform the required operations rapidly and automatically. The boundary condition is defined and a pixel is analyzed and labeled within the one cycle of the pixel clock during which the pixel is held at the P location of the PLW register.

It should be obvious that variations of the disclosed invention are possible in light of these teachings. It is therefore to be understood that, within the scope of the appended claims, the invention may be practiced otherwise than as specifically described.

Patent Citations
Cited PatentFiling datePublication dateApplicantTitle
US3408485 *Feb 24, 1965Oct 29, 1968Perkin Elmer CorpApparatus for counting irregularly shaped objects
US3889234 *Oct 4, 1973Jun 10, 1975Hitachi LtdFeature extractor of character and figure
US4069411 *Jun 24, 1976Jan 17, 1978Bausch & Lomb IncorporatedImage analysis system and method for minimizing paralysis angle
US4115804 *May 23, 1975Sep 19, 1978Bausch & Lomb IncorporatedImage analysis data extraction
US4183013 *Nov 29, 1976Jan 8, 1980Coulter Electronics, Inc.System for extracting shape features from an image
US4189711 *Nov 8, 1977Feb 19, 1980Bell Telephone Laboratories, IncorporatedMultilevel processing of image signals
US4484346 *Apr 28, 1983Nov 20, 1984Sternberg Stanley RNeighborhood transformation logic circuitry for an image analyzer system
US4624013 *Mar 27, 1985Nov 18, 1986Kabushiki Kaisha ToshibaLinked component extraction circuit for image processor
US4630306 *Apr 17, 1984Dec 16, 1986National Research Development Corp.Apparatus and methods for coding and storing raster scan images
US4630307 *Sep 10, 1984Dec 16, 1986Eastman Kodak CompanySignal processing method and apparatus for sampled image signals
Non-Patent Citations
Reference
1"Sequential Operations in Digital Picture Processing", by Asriel Rosenfeld and John L. Pfaltz, Journal of the Association for Computing Machinery, vol. 13, No. 4 (Oct. 1966), pp. 471-494.
2IBM Research Report "An Optimal Component Labelling Algorithm with Label Merging", by Amelia C. Fong, 2/17/84.
3IBM Research Report "On the Reuse of Label Locations in Real Time Component Labelling of Images", by Amelia C. Fong, 12/2/83.
4 *IBM Research Report An Optimal Component Labelling Algorithm with Label Merging , by Amelia C. Fong, 2/17/84.
5 *IBM Research Report On the Reuse of Label Locations in Real Time Component Labelling of Images , by Amelia C. Fong, 12/2/83.
6 *PAL Programmable Array Logic Handbook (Third Edition), by John M. Birkner & Vincent J. Coli, 1983 Monolithic Memories, pp. 1 4 1 16; 4 2 4 11.
7PAL-Programmable Array Logic Handbook (Third Edition), by John M. Birkner & Vincent J. Coli, 1983 Monolithic Memories, pp. 1-4-1-16; 4-2-4-11.
8 *Sequential Operations in Digital Picture Processing , by Asriel Rosenfeld and John L. Pfaltz, Journal of the Association for Computing Machinery, vol. 13, No. 4 (Oct. 1966), pp. 471 494.
Referenced by
Citing PatentFiling datePublication dateApplicantTitle
US4856074 *Mar 8, 1988Aug 8, 1989Fuji Xerox Co., Ltd.Region recognizing device
US4953224 *Feb 16, 1988Aug 28, 1990Hitachi, Ltd.Pattern defects detection method and apparatus
US5014331 *Nov 16, 1989May 7, 1991Fuji Xerox Co., Ltd.Method of detecting an internal point within a closed area
US5164994 *Dec 21, 1989Nov 17, 1992Hughes Aircraft CompanySolder joint locator
US5172423 *Sep 11, 1991Dec 15, 1992Crosfield Electronics Ltd.Methods and apparatus for defining contours in colored images
US5199082 *Nov 1, 1991Mar 30, 1993U.S. Philips Corp.Method of detecting an amplitude transient in a field of elements having a multivalent amplitude distribution, device suitable for performing the method, and video system including the device
US5199083 *Jul 25, 1991Mar 30, 1993Hitachi, Ltd.Image data processing method and system for giving identification labels to areas of connected black picture elements
US5237624 *Aug 16, 1991Aug 17, 1993Fujitsu LimitedReproduction of image pattern data
US5259041 *Nov 16, 1992Nov 2, 1993Konica CorporationMethod of line scanning a document
US5287416 *Aug 17, 1992Feb 15, 1994Unisys CorporationParallel pipelined image processor
US5335294 *Jul 20, 1992Aug 2, 1994Canon Kabushiki KaishaImage processing for pattern extraction
US5420941 *Sep 17, 1993May 30, 1995Unisys CorporationParallel pipelined image processor
US5424823 *Aug 17, 1993Jun 13, 1995Loral Vought Systems CorporationSystem for identifying flat orthogonal objects using reflected energy signals
US5430806 *Sep 7, 1993Jul 4, 1995Loral Vought Systems CorporationSystem for changing perspective of 3-D images obtained from reflected energy signals
US5432619 *Dec 22, 1992Jul 11, 1995Sumitomo Metal Industries, Ltd.Labeling method and apparatus thereof
US5687258 *Sep 3, 1996Nov 11, 1997Eastman Kodak CompanyBorder treatment in image processing algorithms
US6118886 *Mar 30, 1993Sep 12, 2000The United States Of America As Represented By The United States Department Of EnergyAutomatic target recognition apparatus and method
US6483942 *Sep 27, 1999Nov 19, 2002Xerox CorporationMicro region count image texture characterization
US7050625 *Oct 21, 2002May 23, 2006Hewlett-Packard Development Company, L.P.Method and digital camera for indicating when image data has been captured for a three-dimensional target object
US7330288 *Mar 30, 2004Feb 12, 2008Eastman Kodak CompanyPost RIP image rendering in a MICR electrographic printer to improve readability
US7912286 *May 10, 2006Mar 22, 2011Ricoh Company, Ltd.Image processing apparatus and method of image processing capable of effective labeling
US8111919Mar 12, 2008Feb 7, 2012Eyep, Inc.Feature encoding system and method for connected component labeling
US8213734 *Oct 23, 2006Jul 3, 2012Sony Ericsson Mobile Communications AbActive autofocus window
US8249316 *Feb 7, 2006Aug 21, 2012Agency For Science, Technology And ResearchComponent labeling
US8249348Feb 8, 2008Aug 21, 2012Eyep Inc.Label reuse method and system for connected component labeling
US8280167Feb 4, 2008Oct 2, 2012Eyep, Inc.Connected component labeling system and method
US8340421Mar 30, 2008Dec 25, 2012Eyep Inc.Three-dimensional system and method for connection component labeling
US8577143Jul 18, 2012Nov 5, 2013Eyep, Inc.Label reuse method and connected component labeling
US8577144Aug 27, 2012Nov 5, 2013Eyep, Inc.Connected component labeling system and method
US20080007626 *Oct 23, 2006Jan 10, 2008Sony Ericsson Mobile Communications AbActive autofocus window
WO2005084154A2 *Mar 31, 2004Sep 15, 2005Eastman Kodak CoPost rip image
WO2009114609A1 *Mar 11, 2009Sep 17, 2009Eyep, IncFeature encoding system and method for connected component labeling
Classifications
U.S. Classification382/204, 382/266
International ClassificationG06T7/00, G06K9/46
Cooperative ClassificationG06K9/4638, G06T7/00
European ClassificationG06T7/00, G06K9/46A3
Legal Events
DateCodeEventDescription
Feb 25, 1997FPExpired due to failure to pay maintenance fee
Effective date: 19961218
Dec 15, 1996LAPSLapse for failure to pay maintenance fees
Jul 23, 1996REMIMaintenance fee reminder mailed
Feb 4, 1992FPAYFee payment
Year of fee payment: 4
Dec 13, 1991FPAYFee payment
Year of fee payment: 4
Oct 10, 1989CCCertificate of correction